Anthropic launched its new Claude Opus 5 model on Friday, positioning it as a more affordable alternative to its top-tier Fable 5 system.
The model demonstrates strong capabilities in identifying software vulnerabilities.
Internal evaluations based on OSS-Fuzz showed that Opus 5 detects security flaws at a rate nearly equal to the company's Mythos 5 model.
However, it significantly lags behind in generating working exploits.
Safety and Pricing
Anthropic stated that Opus 5 was not trained directly on offensive cyber tasks. Its vulnerability detection abilities stem from broader system improvements.
The company designed safety classifiers for Opus 5 that trigger 85 percent fewer interventions compared to Fable 5.
While Opus 5 can search for vulnerabilities within source code, Anthropic maintains strict blocks against binary scanning, penetration testing, and exploit generation.
Any requests that trigger safety filters automatically fall back to the Opus 4.8 model within Claude platforms.
Participants in Anthropic's Cyber Verification Program can access versions with adjusted restrictions.
Base pricing for Opus 5 remains at $5 per million input tokens and $25 per million output tokens.
